Nov 4, 1985 - mobile laboratory from the Occupational Chest Dis- ease Service of the ..... radiographic correlations, th
British Journal of Industrial Medicine 1986;43:406-413
Pulmonary function in asbestos cement workers: a dose-response study M FINKELSTEIN From the Health Studies Service, Ontario Ministry ofLabour, Toronto, Ontario M7A 117, Canada
This study has found that residence time weighted exposure (asbestos dose) may be used to model the risk and extent of pulmonary function abnormalities in a cohort of asbestos cement workers. This parameter, which incorporates both exposure concentration and latency, had preABSTRACT
viously proved useful for modelling the risk of radiographic abnormalities in this cohort. Asbestos dose and smoking were independent and additive contributors to decreased pulmonary function. It was also found that lung function results could be used as surrogates for dose in the assessment of mortality risk in this cohort. In a previous paper it was noted that asbestos disease may develop or progress after exposure has ceased, and that the usual measure of exposure, the "cumulative exposure," suffers from the shortcoming that its value remains fixed once exposure has ended, requiring that both cumulative exposure and time enter explicitly into risk assessment.' It was shown that an alternative measure called "dose" (which is the sum of the annual asbestos exposure weighted by the retention time) could be used to model the risk of certain radiographic abnormalities in a cohort of asbestos cement workers, and that when using this measure, the assessment of risk had the desirable property of being independent of the time of evaluation over the interval 20-34 years from first exposure. The present study was performed to assess whether "dose" might also be useful in describing the development of pulmonary function abnormalities in this cohort. This measure was found to provide descriptions of the relations between exposure to asbestos and physiological abnormality that were independent of the time of assessment, and it is concluded that dose-that is, the residence time weighted exposuremight be generally useful in describing the relations between exposure to fibrogenic dusts and the risk of disease.
ployed for nine years or more and who had worked at least 12 months in occupations with exposure to asbestos. When six men who had died before the start of pulmonary function testing in 1970 were excluded, 180 men met the criteria for inclusion; of these, 138 (77%) had at least one set of measurements available. Many had been tested several times and 52 had been tested on ten or more occasions. Table 1 gives the numbers of men who had measurements at various intervals from first exposure (latency). After analysis had indicated that the doseresponse pattern was independent of latency a "master" cohort was assembled by selecting one set of measurements from each of the 138 men. The demography of this master cohort is also given in table 1. EXPOSURE AND DOSE ESTIMATION
Information about the factory and the estimation of individual exposures by extrapolation from personal membrane filter measurements has been published.2 Asbestos dosages were calculated by assuming that a fixed proportion of the workplace air concentrations was deposited in the lungs and each year's accumulation was weighted by the residence time in lung tissue (the formula used is given in the appendix). The units for dose are fibres/ml x years squared (f/ml x y2). Materials and methods Cumulative exposures had previously been estimated to be accurate to within a factor of three to STUDY POPULATION five2; since the formula used to calculate dose weights Men eligible for inclusion in the study were pro- early exposures most heavily, the dose estimates used duction workers hired before 1960 who had been em- here are probably more uncertain than the estimates Accepted 4 November 1985 of cumulative exposure. 406
Pulmonary function in asbestos cement workers: a dose-response study Table 1 Study population
407 2------- 3
13
No of men Latency analysis Latency interval: 20-24 years 115 103 25-29 years 30-34 years 46 138 Master cohort Latency interval: 7 15-19 years 41 20-24 years 60 25-29 years 30-34 years 30 Standard deviation 8-5 Mean age: 58 Range 33-78 Standard deviation 6-1 Mean height: 174cm Range: 156-189 Non-smokers: n = 43 Smokers: n = 95
SMOKING INFORMATION
Smoking information had been requested by pulmonary function technicians at the time of examination. Additional information was obtained by questionnaire and from physicians' records. For the purposes of this study, smokers were defined as men who had smoked cigarettes later than 10 years from first exposure; non-smokers were defined as men who had never smoked or who had stopped before 10 years from first exposure. For most of the analyses in this paper, smoking was treated as a dichotomous variable. PULMONARY FUNCTION TESTING (PFT)
All measurements were made either as part of routine surveillance of the workforce or as part of the assessment for workers' compensation claims; none was made specifically for epidemiological purposes. A mobile laboratory from the Occupational Chest Disease Service of the Ontario Ministry of Labour made biannual visits to the plant and performed a standard spirometric evaluation yielding values for forced vital capacity (FVC) and forced expiratory volume in the first second of expiration (FEV1). All volumes were corrected to BTPS. Compensation examinations were performed in the laboratory of the Chest Disease Service and a more elaborate assessment, including the measurement of static lung volumes and exercise testing, was performed. In addition to FVC and FEV1 the tests selected for analysis in this study include total lung capacity (TLC), the single breath carbon monoxide diffusing capacity (uncorrected for haemoglobin concentration) (DLCO), and DLCO per unit lung volume (KCO). The reference equations of Knudson et al and Morris et al were used to make comparisons with populations of healthy non-smoking men.34 Because volumes predicted by these equations frequently differ, comparisons were made with both. The refer-
i
1~
3
3
1
I
-
- "2
'2
LL
I
)OO 300 400 500 6000 7M Dose (fibres/ml x years squared )
800
900
Fig 1 Latency dependence ofFVC against dose. Numbers to left of dashed line are means (with 95% confidence limits) for dose intervals 1200/fml x y2 wide; numbers to right refer to individuals. Latency intervals are: 1 = 20-24 years; 2 = 25-29 years; 3 = 30-34 years. Regression line has been derivedfrom "master cohort" in which each man contributes one value.
ence equation of Miller et al was used for diffusion
tests.5 STATISTICAL METHODS
The relations between PFT results and various explanatory variables were assessed by multiple.linear regression with examination of residuals. Cox proportional hazard regression anaylsis was used- tostudy PFT results as prognostic factors for mortality.6 Results LATENCY DEPENDENCE OF THE DOSE-RESPONSE RELATIONS
The relations between pulmonary function results and the explanatory variables age, height, smoking, and dose were explored by use of multiple linear regression. As noted in table 1 each man provided measurements in one or more latency periods and dose-response relations could be calculated separately for each interval. To test whether these relations differed among latency intervals, all the data were pooled and linear regression analyses were performed using "indicator variables" to specify to which interval each measurement belonged. The results indicated that the temporal terms did not add significantly to the fit of the models; the null hypothesis that the doseresponse relations were the same in all intervals was thus not rejected. This finding is illustrated for FVC in fig 1 in which, for ease of presentation, the data have been grouped in four dose categories. Because the relation between dose and response was effectively independent of the time of assessment,
Finkelstein
408
Table 2 Regression coefficients from multiple linear regression analyses (n = 138). (Parentheses give estimates of the 95% confidence intervals) Constant
Age
Height
Smoking
Dose
R2
-1-77
-0-031 (-0 044, -0-017)
0-044 (0-025, 0.063)
-0-20 (-0 43, +0 03)
-1-7 x 10-4 (-2-5, -1.0)
0-42
(-5-3, + 18)
FEVI (1)
063 (-24, +3 6)
-0-036 (-0048, -0025)
0027 (0011, 0043)
-0-32 (-0-52, -0-13)
-1 1x 10-4 (-1 8, -0 5)
044
FEV,/FVC
0-98 (0-89, 1-07)
-0-0034 (-0-0049, -0-0019)
NS*
-0 05 (-0 077, -0-020)
NS
017
FVC %* (Knudson et al3) (Morris et al4)
986 (92-7, 104 4) 91 4 (86-0, 967)
NS
NS
NS
NS
-5-0 (-10 4, +0 4) -4-6 (-9-6, +04)
-4-0 x 10-3 (-5 7, -2.3) -4-0 x 10-3 (-55, -2.4)
0-16
FEV1 % (Knudson et al) (Morris et al)
978 (91 4,104 3) 101-3 (94-7, 107-8)
NS
NS
NS
-37 x 10-3 (-56, -1-7) -3-8 x 10-3 (-5-7, -1-9)
0-15
NS
-105 (-16.5, -45) -10-8 (-16 9, -4 8)
FVC difft (Knudson et al) (Morris et al)
3-46 (0-35, 6-6) -0 42 (-0-67, -0 18)
NS
-0-02 (-0-038, 0 003) NS
-0-20 (-0 43, +0 02) -0-20 (-0 43, +0 03)
-1-7 x 10-4 (-2-5, -10) -1-7 x 10-4 (-2-4, -1-0)
0-16
FEV, diff (Knudson et al) (Morris et al)
3-63 (0-97, 6-3) NS
-002 (-0 036, -0 006) NS
-0-33 (-0-52, -0 13) -0 32 (-0-52, -0-13)
-1-2 x 10-4 (-1 9, -06) -11 x 10-4 (-1 7, -0 5)
FVC (1)
NS
NS NS
NS = Not significant (p > 0-05). *FVC % = Per cent of predicted from reference equations of Knudson et al3 and Morris et al.' tFVC diff = FVC (observed-predicted): litres.
one set of measurements was chosen from each of the 138 workers who had data available and a pooled "master cohort" was created to provide the largest data set for subsequent analysis. Where possible, the values used were averages over the three consecutive years showing the least variability and with the greatest latency. Since there were relatively few measurements of diffusing capacity, the latency independence of diffusion could not be tested; it was assumed that the dose-response relations for diffusion would also be effectively independent of latency and the master cohort was also used for the analysis of this measure. Three regression models were used to examine the relations between PFT results and the explanatory variables. In the first model the PFT value itself served as the dependent variable; in the two other analyses comparisons were made with the reference equations using proportional (% predicted) and difference (observed-predicted) models. The results for FVC and FEV1 are presented in table 2. Both FVC and FEV1 decreased significantly with increasing dose, about 4% of predicted per 1000 f/ml x y2, and because of the similarity of response their ratio, FEV1/FVC, was independent of dose. Smokers had lower values of both FVC and FEV1 and the contributions of smoking and dose to decreased pulmonary function were additive. This is shown in table
3, which shows that when the regression analyses were run separately for the smokers and non-smokers the dose coefficients were similar. Figure 2 shows the relation between FVC (observed-predicted) and dose; the considerable variability in the data is clearly shown. The diffusion results are presented in table 4. Asbestos dose was associated with significantly lower 2
A
0 0
2 * > LL L ,
AS. .0
O
n0
0
O~~~~~~
Dose (fibres/ml x years squared )
Fig 2 FVC (observed-predicted) against dose for master cohort. Unit for vertical axis is standard error of equation of Knudson et al3 (600 ml) and horizontal line at -165 SE is lower limit of normal. Open circles refer to non-smokers and closedfigures to smokers.
0-14
0-16
0-14
0-17 0-14
409
Pulmonary function in asbestos cement workers: a dose-response study Table 3 Relations between pulmonary function and dose for the smokers and non-smokers.* (Nwnbers in parentheses are estimates of the 95% confidence intervals)
Non-smokers (n = 43) Smokers (n = 95)
FVC
FEV1
FVC %
FEV1 %
TLCt
DLCOt
-14 x 10' (-28, -01) -1-9 x 10-4 (-2-8, -10)
-1-1 X 10-4 (-2-2,0) -1-2 x 10-4 (-2-0, -04)
-3-3 x 10-3 (-6-1, -04) -4-0 x 10-3 (-6-0, -20)
-3-9 x 10-3 (-74, -05) -3-5 x 10-3 (_5-9, -10)
-0-8 x 10-4 (-2-9, +1-3) -2-5 x 10(-4-1, -09)
+1_1 x (-112, -9-9 X (-18-5,
10-4 10-4 -1-3)
+13-4)
*Units of dose are fibres/ml x y2.
tFor TLC and DLCO the number of non-smokers was 26 and the number of smokers was 53. Table 4 Regression coefficients from multiple linear regression analysesfor diffusion
DLCO(ml/min/mmHg) (n = 83) DLCO (% predicted) (Miller) TLC(1) (n = 79) KCO (ml/min/mm Hg/I) (n = 79)
Constant
Age
Height
Smoking
Dose
31-4 (-6-2, +691) 84-3 (75-3, 93-3) -7.5 (-15-0, -0-03) 8-9 (3-2, 14-6)
-014 (-0-26, -001) NS
NS NS
-3-27 (-5-49, -1-05) -11-2 (-19-0, -3-4) NS
-6-7 x 10-4 (-13-6, +02) -1-8 x 10-3 (-4-2, +0 5) -2-2 x 10-4 (-35, -0-9) NS
NS -0-024 (004, -0005)
0079 (0-04, 0-12) NS
NS
R2 0-19 0 11
0-31 0 11
Table 5 Proportion of men with test results less than the lower limit of normal* 0- < 1200 Dose interval Mean dose 840 FVC: unadjusted Knudson et aP1 7/37 (19%) Morris et a!' 11/37 (30%) FVC: smoking adjusted Knudson et al 3/37 (8%) Morris et al 7/37 (19%) FEV1: unadjusted Knudson et al 9/37 (24%) Morris et al 7/37 (19%) FEV1: smoking adjusted Knudson et al 6/37 (16%) Morrisetal 4/37(11%) DLCO: smoking adjusted Miller et al5 4/12 (33%)
*Units for dose are fibres/ml x
1200-
02)
y2.
values for TLC and was of borderline statistical significance for decreased diffusing capacity. Smokers had significantly lower results for DLCO, but not for TLC. Once again, all dose-smoking interaction terms were non-significant. RELATION BETWEEN DOSE AND ABNORMAL PULMONARY FUNCTION
Dose
Fig 3
( fibres
Iml
x years
squared )
Proportion of abnormal values of FVC against dose, adjustedfor smoking, and 5% abnormality rate in reference population. Error bars are estimates of 95% confidence limits.
The dose-response analysis presented in the previous section investigated the relation between dose and pulmonary function each measured on a continuous scale; an alternative approach is to classify each measurement as either normal or abnormal. One method of defining abnormality is by selecting as the lower limit of normal (LLN) the lower 95% confidence limit (or fifth centile) of measurements from healthy, nonsmoking reference populations such as those of Knudson et al or Morris et al. 34 Thus for the present
410 study the LLN was taken to be the predicted value minus 1 645 x the standard error of the reference equation. To assess the relation between dose and the prevalence of abnormalities the men were divided among dose categories 1200 f/ml x y2 wide. As a graphical analysis showed that the rate of abnormality in each interval was essentially independent of latency the master cohort was used for the detailed analysis. To adjust for any smoking differences among dose categories the average volumes lost by smokers, derived from the regression analyses of the previous section, were added to the measured values of the smokers: 0-21 for FVC; 0-32 1 for FEV1 and 3-3 ml/min/mm Hg for DLCO. The results are given in table 5 with, and without, the adjustment for smoking. The reference equations of Knudson et al3 and Morris et al4 predict different LLN and this has an impact on the classification of abnormality. For both FVC and FEV1 the prevalence of abnormalities increased significantly with dose, but an effect could not be shown for DLCO. Tests for departure from a linear trend7 were non-significant. Finney
Finkelstein presents an equation that may be used to adjust the observed proportions for the 5% abnormality rate in the reference populations (see appendix),8 and this adjustment has been used in fig 3 which displays the prevalence of abnormalities of FVC against dose. PULMONARY FUNCTION-RADIOGRAPHIC CORRELATIONS
Radiographs, many of them serial, were available for all but two members of the master cohort and had been interpreted according to the 1971 ILO classification scheme by a NIOSH certified "B" reader before the gathering of the pulmonary function data.9 For the analysis of pulmonary functionradiographic correlations, the film closest in time (within two to three years) to the pulmonary function test used in the master cohort analysis was selected. Men were grouped according to the scores for profusion of small irregular opacities and bilateral pleural thickening and mean values for PFT were computed. The results are presented in table 6 which shows that there was a trend for the average PFT results to worsen as the radiographic appearance
Table 6 Relation between radiographic codes* andpulmonary function results Radiographic code No of men FVC: % Predicted No < LLNt
FEV1/FVC DLCO % Predicted
0/0
0/1
1/0
1/1
> 1/2
75
13
9
31
84-3
(81-0-87-6) 12 (16%)
81 0 (73 5-88 5) 4(31%)
70-5 (606-80 5) 4(44%)
0-76
0-77
0-75 (0-66-0 84)
(0 73-078)
76-2 (n = 32) (698-82-6)
(0-75-079)
8
No pleural thickening 101
Pleural thickening 35
71-9 (66 4-77 5) 14(45%)
63-6 (503-76-8) 4(50%)
81-9 (79-0-84 8) 21 (21%)
70-8 (653-76-3) 17 (49%)
0-73 (0 69-0 76)
0-77 (0 70-0-83)
0 75 (0-73-0 77)
0-76
55-3 (n = 8) (450-65 5)
75 0 (n = 54) (69 7-80 4)
65-5 (n = 29) 600-71-1)
7 (88%)
20 (37%)
18 (62%)
74-7 (n = 8)
75-1 (n = 9) 69-2 (n = 26) (608-89 4) (623-76-1) < No LLN 12 (38%) 4 (50%) 4 (44%) 12 (46%) *Radiographic codes are for small irregular opacities and bilateral pleural thickening. tLLN is the lower limit of normal: 1-65 standard errors below the predicted value.
(54.2-95.3)
(0-730-79)
Table 7 Results of Cox regression mortality analysisfor 115 men entering observation at 20-24 years latency Variable
FVC: (fl)t FEV1: (f)t FEV1/FVC
Xray >1/1 Multivariate models$
Cause of death All causes (n = 40)
Mesothelioma (n = 9)
p < 0-001 (-0-50) p < 0 001 (-0.50)
p < 0-05 p < 0-10
NS
(-0.57) (-0 42)
p < 0-001
NS NS
FVC (p < 0 04) Xray(p < 0-01)
Age (p < 0 04) FVC(p < 007)
Lung cancer (n = 11) p < 0 05 (-0 57) p
